Hierarchical Junction Trees as the Secondary Structure for Inference in Bayesian Networks

Traditionally, a single junction tree is used as the secondary structure for inference in a Bayesian network. However, its applicability and efficiency are restricted by the size of the junction tree. In this paper, we demonstrate that using a hierarchy of junction trees (HJT) as the secondary structure instead will greatly alleviate this restriction and improve the performance. We also compare the proposed HJT with other similar schemes for inference in Bayesian networks.